Closed zhangjiulong closed 8 years ago
Solved as https://github.com/srvk/eesen/issues/78
Hi, I am facing the same problem while setting up the cluster. Using "qstat -j 29" checking the error as following:
script_file: /home/gpu/software/kaldi-20181120/egs/thchs30/s5/exp/make_mfcc/train/q/make_mfcc_train.sh job-array tasks: 1-8:1 error reason 4: 01/07/2019 15:19:42 [0:14111]: error: can't open output file "/home/gpu/software/kaldi-20181120/egs/ error reason 5: 01/07/2019 15:19:42 [0:14112]: error: can't open output file "/home/gpu/software/kaldi-20181120/egs/ error reason 6: 01/07/2019 15:19:42 [0:14113]: error: can't open output file "/home/gpu/software/kaldi-20181120/egs/ error reason 7: 01/07/2019 15:19:42 [0:14114]: error: can't open output file "/home/gpu/software/kaldi-20181120/egs/ error reason 8: 01/07/2019 15:19:42 [0:14115]: error: can't open output file "/home/gpu/software/kaldi-20181120/egs/ scheduling info: Job is in error state
I had checked the output file. It exists. It looks like a permission issue. But I had use "chown -R" to change the permission, it still had the same error. Could you please guide me on how to solve the problem. Thx.
try root
I had solved by adding "no_root_squash" in /etc/exports file : /home/gsadmin * (rw,sync,no_subtree_check,no_root_squash)
Thank you all.
Hi I get a gridengine cluster which has three nodes(node1,node2,node3), each node has 3 gpus. node1 is master node, submit node and exector node node2 is submit node and executor node node3 is submit node Also node3 has a nfs service, all the wav data and txt data is on node3. Node1 and node2 mount node3's data and node1 and node2 can only read, can not write the mounted data.
node1, node2 and node3 has the same user name called kaldi and password
Then on node1, node2 and node3's dir named /home/kaldi/git/eesen I built eesen separately. I touched uname.sh contains uname -a and runned on node3(only submit node) several times, the job was distributed to node1 and node2 and runs ok
Then I touched a file named touch.sh which contains 'touch ok.fst' cmd and submited on node3 several times, the job was distributed to node1 and node2 and runs ok
Then I went to timit dir and changed the cmd.sh to queue.pl on the three nodes On node3 I runned run_ctc_phn.sh, then runed qstat -j 54, I got the following result:
then I went to node 1 and runed run_ctc_phn.sh and got the similar result half of the cmd runed ok:
Then I disabled node2 and runed run_ctc_phn.sh on node1 (the cluster now has only one exector node1) then eesen runed ok
What is the problem? Please help me to find out how the problem comes? Thanks.